Computer-use traces are a rich source of knowledge about how everyday work gets done. They could help us document, audit, and reuse expertise that is otherwise lost.
Modeling these traces poses three challenges:
1. raw traces consist of keyboard and mouse activities that carry little semantic meaning
2. real work is multi-threaded, with goals interleaved within a session
3. a trace shows the executed path, which does not directly reveal either the goal hierarchy that motivated each step or the control flow that organized the execution.
Our paper introduces Task Model Induction (TMI), turning traces into symbolic, reusable task models. (EMNLP 2026)
14K Followers 199 FollowingAssistant Prof at Stanford CS, member of @stanfordnlp and statsml groups; Formerly at Microsoft / postdoc at Stanford CS / Stats.
28K Followers 302 FollowingProfessor of linguistics and professor of computer science at Stanford and author of the James Beard award finalist "The Language of Food"
4K Followers 475 FollowingMax Palevsky Professor of Sociology, Computational & Data Science @UChicago, Santa Fe Institute, & Google tweeting about science, technology, and AI in society.
6K Followers 1K FollowingKellogg Chair of Technology at Kellogg, Founding director, Center for Science of Science and Innovation, Northwestern University
19K Followers 2K Following@stanford Professor of Computer Science, @simile_ai co-founder, nationally bestselling author. I build interactive, social, and societal tech.